A method, device, terminal and storage medium for slope stability assessment

By combining the SSA-XGBoost classification model with the geological weighted loss function and the FocalLoss function, the slope stability assessment is optimized, which solves the problems of high computational cost and model omission in traditional methods, and realizes efficient and accurate assessment of slopes for power facilities.

CN123069877APending Publication Date: 2026-11-10ZHONGSHAN POWER SUPPLY BUREAU OF GUANGDONG POWER GRID
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611235217.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-14
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Traditional slope stability assessment methods are computationally expensive and inefficient, making them difficult to apply rapidly around large areas of power facilities. Furthermore, assessment methods based on the XGBoost model tend to favor the majority class in imbalanced data, resulting in insufficient sensitivity in identifying potentially dangerous slopes and the possibility of missing instability risks.

Method used

The SSA-XGBoost classification model is adopted, which combines the geological weighted loss function and the FocalLoss function. By introducing geological complexity weights and dynamically adjusting the focusing parameters, the model training process is optimized, thereby improving the ability to identify unstable slopes.

Benefits of technology

It effectively solves the problem of category imbalance, improves the ability to identify slopes with complex geological structures, reduces the risk of missed detection, provides accurate stability assessment results, and provides reliable technical support for the safe construction and operation and maintenance of power facilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069877A_ABST
    Figure CN123069877A_ABST
Patent Text Reader

Abstract

This application discloses a slope stability assessment method, device, terminal, and storage medium, relating to the field of slope stability assessment technology. The solution provided in this application introduces a weighted loss function based on geological complexity weights and the FocalLoss function, enabling it to dynamically focus on a small number of critical unstable slope samples. This effectively solves the common class imbalance problem in power facility slope stability assessment and improves the ability of the SSA-XGBoost classification model to identify slopes with complex geological structures. This method can utilize multi-source slope feature data for model training and output accurate stability assessment results, avoiding the risk of missed detection of potentially dangerous slopes by existing machine learning models, thus providing reliable technical support for the safe construction and operation of power facilities.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of slope stability assessment technology, and in particular to a slope stability assessment method, device, terminal and storage medium. Background Technology

[0002] With the continuous expansion of my country's power grid, a large number of power facilities such as transmission line towers, substations, and wind turbines have to be located in mountainous, hilly, or high-fill areas with complex terrain. These sites often face slopes, whether artificially excavated or naturally formed, and their stability directly affects the structural safety and operational reliability of the power facilities. Once a slope experiences instability or collapse, it can lead to anything from tower tilting and line tripping to tower collapse, line breakage, power outages, and even significant economic losses. Therefore, accurate assessment of slope stability has become a key technical requirement in the construction and operation and maintenance phases of power engineering projects.

[0003] Traditional slope stability assessment methods mainly rely on limit equilibrium methods and finite element numerical simulations. While theoretically mature, these methods typically require detailed geological survey parameters and complex mechanical modeling, resulting in high computational costs and low efficiency, making them difficult to rapidly apply across large areas and multiple measurement points around power facilities. In recent years, with the development of machine learning technology, slope stability assessment methods based on ensemble learning models such as XGBoost have demonstrated good nonlinear mapping capabilities and assessment accuracy. However, in power facility construction scenarios, slope instability cases are often low-probability events, leading to a severe class imbalance in the training data—stable slope samples far outnumber unstable samples. Conventional XGBoost models, when faced with such imbalanced data, tend to favor the majority class, reducing the sensitivity to identify potentially dangerous slopes and thus leading to missed detections of slope instability risks. Summary of the Invention

[0004] This application provides a slope stability assessment method, device, terminal, and storage medium to achieve the invention objective of improving the sensitivity of identifying unstable slopes and reducing the risk of missed detection.

[0005] To achieve the above-mentioned objectives, the first aspect of this application provides a slope stability assessment method, comprising:

[0006] Obtain slope feature data samples, where each set of slope feature data samples includes: slope geometry, geotechnical parameters, and environmental dynamic parameters;

[0007] Based on the preset SSA-XGBoost classification model, the slope feature data samples are used as input features of the SSA-XGBoost classification model. The model is trained by combining the geological weighted loss function built into the SSA-XGBoost classification model. The geological weighted loss function is a weighted function constructed based on geological complexity weights and FocalLoss function. The geological complexity weights are used to quantify the geological structural complexity of the slope samples.

[0008] Based on the calculation results of the geological weighted loss function, the convergence degree of the SSA-XGBoost classification model is determined. When the convergence degree or the number of iterations meets the preset training termination condition, the slope stability assessment model is obtained.

[0009] Obtain slope characteristic data of the target area, input the slope characteristic data into the slope stability assessment model, and obtain the slope stability assessment result of the target area through the calculation of the slope stability assessment model.

[0010] Preferably, before obtaining the slope stability assessment result of the target area through the calculation of the slope stability assessment model, the method further includes:

[0011] The rate of change of the environmental dynamic parameters is detected, and the focus parameters in the geological weighted loss function are dynamically updated based on the preset correspondence between the parameter change rate and the focus parameters.

[0012] Preferably, the step of detecting the rate of change of the environmental dynamic parameters and dynamically updating the focused parameters in the geological weighted loss function in conjunction with a preset correspondence between the parameter change rate and the focused parameters includes:

[0013] When the rate of change of the environmental dynamic parameters is greater than the preset rate of change threshold, the focus parameter in the geological weighted loss function is increased according to the preset correspondence between the parameter rate of change and the focus parameter. When the rate of change of the environmental dynamic parameters is not greater than the preset rate of change threshold, the focus parameter in the geological weighted loss function is decreased according to the preset correspondence between the parameter rate of change and the focus parameter.

[0014] Preferably, the SSA-XGBoost classification model is constructed in the following ways:

[0015] Based on each feature parameter included in the slope feature data sample, a global sensitivity analysis is performed on the feature parameter using the Morris screening method or the Sobol index method to determine the sensitivity of each feature parameter to slope stability, and based on the sensitivity, the feature parameter is divided into sensitive parameters and insensitive parameters.

[0016] Based on the sensitivity corresponding to each feature parameter, and combined with the preset correspondence between parameter sensitivity and Levy flight step size, the Levy flight step size corresponding to each feature parameter is determined, wherein the Levy flight step size corresponding to the sensitive parameter is smaller than the Levy flight step size corresponding to the non-sensitive parameter.

[0017] Based on the slope characteristic data sample, the sparrow population is initialized according to the typical parameter range corresponding to the typical slope instability mode;

[0018] Based on the sparrow population and the correspondence between feature parameters and Levy flight stride length, the initial XGBoost classification model is hyperparameter optimized according to the optimization logic of the SSA optimization algorithm, resulting in the SSA-XGBoost classification model.

[0019] Preferably, the step of using the slope feature data samples as input features of the SSA-XGBoost classification model based on the preset SSA-XGBoost classification model further includes:

[0020] Based on the slope feature data samples, the stability coefficient corresponding to each set of slope feature data samples is calculated using the preset limit equilibrium strip method, so as to use the stability coefficient as the prior input feature of the SSA-XGBoost classification model.

[0021] Preferably, the limit balance slice method includes: the simplified Bishop method or the unbalanced thrust method.

[0022] Preferably, the step of using the slope feature data samples as input features of the SSA-XGBoost classification model based on the preset SSA-XGBoost classification model further includes:

[0023] Based on the lithology type corresponding to the slope feature data samples, when the lithology type similarity of any two sets of unstable samples is higher than a preset similarity threshold, the two sets of unstable samples are combined into a sample pair.

[0024] Interpolation is performed on each sample pair to generate a synthetic sample, which is used to expand the slope feature data sample.

[0025] The second aspect of this application provides a slope stability assessment apparatus for performing the slope stability assessment method as provided in the first aspect of this application, including:

[0026] The sample acquisition unit is used to acquire slope feature data samples. Each set of slope feature data samples includes: slope geometry, soil and rock parameters, and environmental dynamic parameters.

[0027] The evaluation model training unit is used to train the model based on the preset SSA-XGBoost classification model, taking the slope feature data samples as input features of the SSA-XGBoost classification model, and combining the geological weighted loss function built into the SSA-XGBoost classification model. The geological weighted loss function is a weighted function constructed based on geological complexity weights and the FocalLoss function. The geological complexity weights are used to quantify the geological structural complexity of the slope samples.

[0028] The model convergence determination unit is used to determine the convergence degree of the SSA-XGBoost classification model based on the calculation results of the geological weighted loss function. When the convergence degree or the number of iterations meets the preset training termination condition, the slope stability assessment model is obtained.

[0029] The slope stability measurement and evaluation unit is used to acquire slope characteristic data of the target area, input the slope characteristic data into the slope stability evaluation model, and obtain the slope stability evaluation result of the target area through the calculation of the slope stability evaluation model.

[0030] A third aspect of this application provides a slope stability assessment terminal, comprising: a memory and a processor;

[0031] The memory is used to store program code, which corresponds to the slope stability assessment method provided in the first aspect of this application.

[0032] The processor is used to read and execute the program code to implement the slope stability assessment method.

[0033] The fourth aspect of this application provides a computer-readable storage medium storing program code that is read and executed by a processor to implement the slope stability assessment method provided in the first aspect of this application.

[0034] As can be seen from the above technical solutions, this application has the following advantages:

[0035] The proposed solution introduces a weighted loss function based on geological complexity weights and the FocalLoss function, enabling it to dynamically focus on a small number of critical unstable slope samples. This effectively addresses the common class imbalance problem in power facility slope stability assessment and enhances the SSA-XGBoost classification model's ability to identify slopes with complex geological structures. This method utilizes multi-source slope feature data for model training and outputs accurate stability assessment results, avoiding the risk of missed detections of potentially dangerous slopes by existing machine learning models. This provides reliable technical support for the safe construction and operation of power facilities. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a schematic flowchart illustrating an embodiment of a slope stability assessment method provided in this application.

[0038] Figure 2 This is a schematic diagram of the overall process of an embodiment of a slope stability assessment method provided in this application.

[0039] Figure 3 This is a schematic diagram of the architecture of an embodiment of a slope stability assessment device provided in this application. Detailed Implementation

[0040] With the continuous expansion of my country's power grid, a large number of power facilities are located in mountainous, hilly, or high-fill areas with complex terrain. The stability of their slopes directly affects the safety and operational reliability of these facilities. Traditional slope stability assessment methods are computationally expensive and inefficient, making them difficult to apply rapidly to power facilities across large areas and at multiple measurement points. Existing machine learning-based XGBoost models, when faced with the low-probability event of slope instability at power facilities, suffer from insufficient sensitivity in identifying potentially dangerous slopes due to imbalanced training data, tending to favor the majority class and resulting in missed instability risks.

[0041] In view of this, embodiments of this application provide a slope stability assessment method, apparatus, terminal, and storage medium to achieve the invention objective of improving the sensitivity of identifying unstable slopes and reducing the risk of missed detection.

[0042] To make the inventive objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0043] First, a detailed description of an embodiment of a slope stability assessment method provided in this application is as follows:

[0044] Please see Figure 1This application provides an embodiment of a slope stability assessment method, the steps of which include:

[0045] Step 101: Obtain slope feature data samples;

[0046] Step 102: Based on the preset SSA-XGBoost classification model, the slope feature data samples are used as input features of the SSA-XGBoost classification model, and the model is trained by combining the geological weighted loss function built into the SSA-XGBoost classification model.

[0047] Step 103: Based on the calculation results of the geological weighted loss function, determine the convergence degree of the SSA-XGBoost classification model. When the convergence degree or the number of iterations meets the preset training termination condition, the slope stability assessment model is obtained.

[0048] Step 104: Obtain slope characteristic data of the target area, input the slope characteristic data into the slope stability assessment model, and obtain the slope stability assessment result of the target area through the calculation of the slope stability assessment model.

[0049] For ease of understanding, the following explains some key terms in this embodiment:

[0050] Slope characteristic data samples refer to historical data sets used to train and validate slope stability assessment models. Each set of samples contains multiple dimensions of information describing the slope's condition, such as the slope's geometry, the physical and mechanical properties of the soil or rock mass constituting the slope, and external environmental factors acting on the slope. These data are acquired through various methods, including field surveys, laboratory experiments, and real-time monitoring.

[0051] The SSA-XGBoost classification model is an ensemble learning model that combines the Sparrow Search algorithm with gradient boosting trees. XGBoost is an efficient, flexible, and scalable gradient boosting framework that exhibits excellent performance in handling structured data. SSA is a novel swarm intelligence optimization algorithm used to optimize the hyperparameters of the XGBoost model, thereby improving its global search capability and classification performance.

[0052] The geologically weighted loss function is a function used during the training of a machine learning model to measure the difference between the model's predictions and the true labels. This function calculates the prediction error and guides the iterative update of the model parameters, gradually converging them to their optimal state. In this embodiment, the geologically weighted loss function is designed as a weighted function to address the challenges posed by class imbalance and geological complexity in slope stability assessment.

[0053] Geological complexity weight is a numerical factor used to quantify the geological structural complexity of slope samples. This weight is comprehensively evaluated based on factors such as lithological combination, structural plane development, fault distribution, and hydrogeological conditions of the slope. The geological complexity weight is incorporated into the geological weighted loss function, enabling the model to pay greater attention to slope samples with complex geological conditions during training, thereby improving the model's ability to identify the instability risk of complex slopes.

[0054] The FocalLoss function is an improved cross-entropy loss function, primarily used to address class imbalance problems in tasks such as object detection. This function introduces a focusing parameter, reducing the weight of easily classified samples, causing the model to focus more on difficult-to-classify samples, especially minority class samples. In this embodiment, the FocalLoss function is used to handle the difference in the number of stable and unstable samples in slope stability assessment.

[0055] The slope stability assessment model refers to a trained and converged SSA-XGBoost classification model. This model can take new slope feature data as input and output the stability assessment result of the slope, such as determining whether the slope is stable, basically stable, or unstable. This model is the core tool for practical engineering applications in this embodiment.

[0056] Slope geometry refers to the external morphological characteristics of a slope, including slope height, slope angle, and slope curvature. These parameters are obtained through technologies such as 3D laser scanning and are used to describe the macroscopic geometric features of the slope.

[0057] Geotechnical parameters refer to the physical and mechanical properties of the soil or rock that constitutes a slope, including cohesion, angle of internal friction, unit weight, and permeability coefficient. These parameters are determined through methods such as in-situ direct shear tests and indoor triaxial tests, reflecting the strength and deformation characteristics of the slope materials.

[0058] Environmental dynamic parameters refer to external environmental factors that affect slope stability, including groundwater level, rainfall, and seismic motion parameters. These parameters are collected in real time through an Internet of Things (IoT) sensor network, reflecting the dynamic changes in the environment in which the slope is located.

[0059] Specifically, the slope stability assessment method in this embodiment can be implemented in the following manner:

[0060] First, slope characteristic data samples are obtained. This process may include collecting slope geometry, geotechnical parameters, and environmental dynamic parameters through methods such as manual field surveys, reviewing historical geological reports, and conducting routine indoor geotechnical tests. For example, slope height, slope angle, and other geometric features can be measured manually or extracted from existing topographic maps; geotechnical parameters such as cohesion and internal friction angle can be obtained by conducting standard direct shear tests or triaxial tests on the slope soil; environmental dynamic parameters such as groundwater level and rainfall can be obtained from historical records of local meteorological or hydrological stations. Each set of data samples is formatted into a standardized format for subsequent model training.

[0061] Secondly, based on the pre-defined SSA-XGBoost classification model, the slope feature data sample is used as the input feature of the SSA-XGBoost classification model, and the model is trained using the built-in geological weighted loss function. During this process, a pre-built SSA-XGBoost classification model is loaded. The slope feature data sample, including its geometric structure, soil and rock parameters, and environmental dynamic parameters, is directly input into the model. The purpose of model training is to iteratively optimize its internal parameters so that it can accurately classify slope stability. This optimization process replaces the standard XGBoost loss function with a pre-defined geological weighted loss function, which is a weighted function constructed based on geological complexity weights and the FocalLoss function. The specific expression can be found in:

[0062] ;

[0063] Among them, the geological complexity weight These are fixed values ​​predetermined based on the geological complexity of the slope samples. They are used in the loss function to assign different levels of attention to samples with varying degrees of complexity. For example, areas with faults, dense joints, and lithological contact zones are given higher weights. This refers to the FocalLoss function.

[0064] The expression for the FocalLoss function can be found here:

[0065] ;

[0066] in, This is a dynamic scaling factor that automatically reduces the weight of "easily classified samples" (a large number of stable slopes), allowing the model to focus on "difficult-to-classify samples" (a small number of unstable slopes and boundary samples). The model's predicted probability for the correct class: when When the value is close to 1 (easily classifiable samples), the factor approaches 0, reducing its contribution to the loss; when Samples close to 0 (difficult to classify) have factors close to 1, and high loss weights are retained.

[0067] These are the weights for positive samples, used to balance the overall weights of positive and negative samples.

[0068] The focusing parameter is used to amplify the loss gradient of hard-to-classify samples.

[0069] By using the geological weighted loss function mentioned above, the model is forced to prioritize learning patterns in high-risk and geologically complex areas. Engineering risk assessment experience is integrated into the model training, and the model's attention ratio to easily classified and difficult-to-classify samples is optimized to address the class imbalance problem.

[0070] Next, based on the calculation results of the geologically weighted loss function, the convergence degree of the SSA-XGBoost classification model is determined. In each iteration of model training, the geologically weighted loss function value under the current model parameters is calculated. By observing the trend of this loss function value with the number of iterations, the learning progress of the model can be judged. When the decrease in the loss function value is less than a preset threshold, it indicates that the improvement in model performance is leveling off, or when the number of training iterations reaches a preset maximum value, the training process is terminated. At this point, the model parameters are considered to have converged to the optimal state, thus obtaining a slope stability assessment model that can be used for practical evaluation.

[0071] Finally, slope characteristic data for the target area is acquired and input into the slope stability assessment model. The model then calculates the slope stability assessment results for the target area. Specifically, for a specific slope area requiring stability assessment, its current geometric features, geotechnical parameters, and environmental dynamic parameters are collected. For example, geometric features can be obtained through manual measurement or drone aerial photography; geotechnical parameters can be estimated based on field surveys or empirical data; and environmental dynamic factors can be obtained from real-time monitoring equipment or recent meteorological data. This data is processed and input into the pre-trained slope stability assessment model. The model processes and calculates the input data, ultimately outputting the stability status of the slope in the target area, such as a classification result like "stable," "basically stable," or "instable."

[0072] This embodiment effectively addresses the common class imbalance problem in power facility slope stability assessment by introducing a weighted loss function based on geological complexity weights and the FocalLoss function, and improves the model's ability to identify slopes with complex geological structures. This method can utilize multi-source slope feature data for model training and output accurate stability assessment results, thus overcoming the limitations of high computational cost and low efficiency of traditional methods. It also avoids the risk of existing machine learning models missing potential dangerous slopes, providing reliable technical support for the safe construction and operation of power facilities.

[0073] In some specific embodiments, this application further proposes that, before obtaining the slope stability assessment results for the target area through the calculation of the slope stability assessment model, the following steps are also included:

[0074] The rate of change of dynamic environmental parameters is detected, and the focus parameters in the geological weighted loss function are dynamically updated based on the preset correspondence between the rate of change of parameters and the focus parameters.

[0075] Specifically, detecting the rate of change of dynamic environmental parameters refers to real-time monitoring of dynamic environmental parameters such as groundwater level, rainfall, and seismic motion parameters in the slope area, and calculating the magnitude or rate of change of these parameters within a specific time window. For example, it can calculate the rainfall intensity per unit time, or the rate of rise / fall of the groundwater level. This is typically achieved by collecting data in real time through an IoT sensor network deployed in the slope area, and then performing periodic calculations and analyses by a data processing module.

[0076] Among them, the focusing parameters This is a key parameter in the geologically weighted loss function (a weighted function constructed based on geological complexity weights and the FocalLoss function), typically used to adjust the model's focus on difficult-to-classify samples (especially misclassified samples). The preset correspondence between the parameter change rate and the focus parameter refers to establishing a mapping rule or lookup table that defines the focus parameter values ​​corresponding to different ranges of environmental dynamic parameter change rates. For example, when rainfall intensity or groundwater level rise is detected to be relatively low, the focus parameter... A smaller value can be chosen to make the model pay relatively balanced attention to all samples; while when the rate of change is large, the focus parameter can be adjusted. A larger value can be selected to enhance the model's sensitivity to these dynamic changes, making it pay more attention to samples that may be in a critical state in the current environment.

[0077] Dynamically updating the focusing parameters in the geologically weighted loss function means that during the actual operation of the slope stability assessment model, the focusing parameters in the FocalLoss function are adjusted according to the real-time detected rate of change of environmental dynamic parameters, based on the aforementioned pre-defined correspondence. This dynamic adjustment allows the model to adjust its focus on sample classification in real time based on the urgency of the current environment.

[0078] Through the aforementioned technical solution, the slope stability assessment model can dynamically adjust the focusing parameters in its geological weighted loss function based on the real-time rate of change of environmental dynamic parameters. When environmental dynamic parameters change drastically, the model enhances its sensitivity to critical states or boundary samples, thereby enabling faster identification of potential instability risks in early warning scenarios. This "dynamic weight adjustment based on monitoring data stream" mechanism allows the model to adjust its sensitivity to boundary samples according to real-time rainfall, water level, and other data, effectively avoiding assessment lag caused by sudden environmental changes and significantly improving the timeliness and accuracy of early warnings.

[0079] In some of the above embodiments, this application further proposes a step of detecting the rate of change of the environmental dynamic parameters and dynamically updating the focused parameters in the geological weighted loss function by combining a preset correspondence between the parameter change rate and the focused parameters, including:

[0080] When the rate of change of the environmental dynamic parameters is greater than the preset rate of change threshold, the focus parameter in the geological weighted loss function is increased according to the preset correspondence between the parameter rate of change and the focus parameter. When the rate of change of the environmental dynamic parameters is not greater than the preset rate of change threshold, the focus parameter in the geological weighted loss function is decreased according to the preset correspondence between the parameter rate of change and the focus parameter.

[0081] Specifically, the rate of change of environmental dynamic parameters refers to the magnitude of numerical changes in environmental dynamic parameters such as groundwater level, rainfall, or seismic motion parameters within a certain time window, such as the past hour, day, or several days. This rate of change can be obtained by calculating the data difference between adjacent time points, the moving average, or by using time series analysis methods (such as differential operations). Its function is to quantify the dynamic intensity and trend of external environmental loads on slopes, providing a basis for judging the urgency of slope stability risks. For example, the rate of change of rainfall can reflect the intensity of short-term heavy rainfall or continuous rainfall, and the rate of change of groundwater level can indicate a rapid rise or fall in groundwater level. A preset rate of change threshold is a pre-set critical value used to distinguish whether changes in environmental dynamic parameters fall within the normal fluctuation range or are significant changes that may trigger slope instability. This threshold can be determined based on historical slope monitoring data, regional hydrogeological characteristics, engineering experience, safety specifications, or through risk assessment models. For example, a rainfall rate of change threshold can be set; when the actual rainfall rate of change exceeds this threshold, it indicates a potentially high risk of landslides. Parameter change rate and focused parameters The correspondence defines how the rate of change of environmental dynamic parameters affects the focusing parameter in the geological weighted loss function. This can be a function mapping, a lookup table, or a series of conditional judgment rules. For example, it can be set that the focusing parameter takes a specific value when the rate of change is within a certain range; or the focusing parameter has a linear or non-linear relationship with the rate of change. This correspondence aims to transform the dynamic changes of environmental dynamic parameters into the model's attention to the difficulty of sample classification, thereby achieving adaptive adjustment of model sensitivity. When the rate of change of the environmental dynamic parameters is greater than a preset rate of change threshold, increasing the focusing parameter in the geological weighted loss function means that the model will pay more attention to samples that are difficult to classify correctly, especially those samples that are in a critical state between stability and instability. This enables the model to more sensitively capture potential signs of instability when the external environmental conditions of the slope deteriorate, thereby improving the timeliness and accuracy of early warning. Conversely, when the rate of change of the environmental dynamic parameters is not greater than the preset rate of change threshold, decreasing the focusing parameter in the geological weighted loss function indicates that the model's attention to easily classified samples is relatively increased, while its attention to difficult-to-classify samples is relatively decreased. This helps to reduce the model's overreaction to minor fluctuations when the external environmental conditions of the slope are relatively stable or improving, thereby reducing the false alarm rate, improving the robustness and stability of the model, and avoiding unnecessary warnings.

[0082] Through the above technical solution, this application can adaptively adjust the sensitivity of the slope stability assessment model according to the real-time changes of environmental dynamic parameters. Specifically, when the rate of change of environmental dynamic parameters (such as rainfall, groundwater level, or seismic motion parameters) exceeds a preset threshold, indicating a significant increase in external loads or environmental risks faced by the slope, by increasing the focusing parameter in the geological weighted loss function, the model will focus more on learning samples that are in a critical state or difficult to distinguish, thereby improving the model's ability to identify potential instability risks and the timeliness of early warning. Conversely, when the rate of change of environmental dynamic parameters is within the normal fluctuation range or below the preset threshold, by reducing the focusing parameter, the model will reduce its sensitivity to small fluctuations, reduce excessive attention to easily classified samples, effectively avoid false alarms caused by normal fluctuations in environmental parameters, and improve the stability and reliability of the model. This dynamic adjustment mechanism enables the slope stability assessment model to intelligently adjust its early warning strategy according to the actual environmental risk level, ensuring rapid response in high-risk scenarios while also guaranteeing the accuracy and robustness of assessment in low-risk scenarios, thereby significantly improving the effectiveness of slope early warning.

[0083] In some specific embodiments, this application further proposes a method for constructing the SSA-XGBoost classification model, specifically including:

[0084] Based on each feature parameter included in the slope feature data sample, a global sensitivity analysis is performed on the feature parameters using the Morris screening method or the Sobol index method to determine the sensitivity of each feature parameter to slope stability. Based on the sensitivity, the feature parameters are divided into sensitive parameters and insensitive parameters. According to the sensitivity of each feature parameter, and combined with a preset correspondence between parameter sensitivity and Levy flight step length, the corresponding Levy flight step length is determined, wherein the Levy flight step length corresponding to the sensitive parameter is smaller than the Levy flight step length corresponding to the insensitive parameter. Based on the slope feature data sample, multiple sparrow populations are initialized according to the typical parameter range corresponding to typical slope instability modes. Based on the sparrow populations, and combined with the correspondence between feature parameters and Levy flight step length, the initial XGBoost classification model is hyperparameter optimized according to the optimization logic of the SSA optimization algorithm, resulting in an SSA-XGBoost classification model.

[0085] The Morris screening method or the Sobol index method are used to conduct global sensitivity analysis on characteristic parameters, aiming to quantify the impact of each input characteristic parameter on the slope stability assessment results. The Morris screening method is a qualitative global sensitivity analysis method that assesses the sensitivity of each input parameter by calculating the mean and standard deviation of its "basic effect." Its computational cost is relatively low, making it suitable for the initial screening of important parameters. The Sobol index method is a quantitative global sensitivity analysis method based on variance decomposition, capable of calculating the first-order sensitivity index and the total sensitivity index, providing a more accurate quantification of sensitivity. These methods can identify "sensitive parameters" that significantly affect slope stability and "non-sensitive parameters" with less impact, providing guidance for subsequent optimization processes.

[0086] In determining the Levy step size for each feature parameter, this application utilizes the results of global sensitivity analysis. Levy walking is a random walk strategy with a heavy-tailed step size distribution, which effectively enhances the global search capability of the optimization algorithm and avoids getting trapped in local optima. Here, the Levy step size for sensitive parameters is set smaller than that for non-sensitive parameters. This means that for critical parameters that have a significant impact on slope stability, the optimization algorithm will perform a more refined and localized search to accurately find their optimal value; while for non-critical parameters with a smaller impact, a wider range of exploration is allowed to improve search efficiency. This adaptive step size adjustment strategy allows the optimization process to balance the breadth and depth of the search.

[0087] Initializing multiple sparrow populations is one of the key steps in the Sparrow Search Algorithm (SSA). This application initializes sparrow populations based on slope characteristic data samples and the typical parameter ranges corresponding to typical slope instability modes. This means that at the start of optimization, the sparrow populations (i.e., candidate hyperparameter combinations) are set within a physically meaningful parameter space related to the actual slope instability situation. For example, typical value ranges for slope geometry, soil parameters, and environmental dynamic parameters under different instability modes (such as sliding, toppling, creep, etc.) can be determined based on historical instability cases or expert experience. This initialization method ensures that the optimization process starts from a more promising region, thereby accelerating convergence and improving the quality of the final optimization results.

[0088] Subsequently, based on the sparrow population and the correspondence between feature parameters and Levy flight step length, the initial XGBoost classification model was hyperparameter optimized according to the optimization logic of the SSA optimization algorithm. The SSA optimization algorithm simulates the foraging and anti-predation behavior of sparrows, finding the optimal solution by iteratively updating the sparrow's position (i.e., the hyperparameter values ​​of the XGBoost model). In each iteration, the sparrow's movement step length is guided by the Levy flight mechanism, and this step length is adaptively adjusted according to the aforementioned parameter sensitivity. Small adjustments to sensitive parameters help to accurately approximate the optimal solution, while larger adjustments to insensitive parameters help to escape local optima and explore a broader hyperparameter space. In this way, the SSA algorithm can efficiently search for the optimal hyperparameter combination of the XGBoost model, thus obtaining a highly optimized SSA-XGBoost classification model for slope stability assessment tasks.

[0089] Through the above technical solutions, this application effectively solves the model optimization problem caused by the non-convexity of the slope parameter space. Global sensitivity analysis can identify key parameters affecting slope stability and adaptively adjust the Levy flight step size accordingly, enabling the optimization algorithm to conduct targeted exploration in the hyperparameter space. For sensitive parameters, a smaller Levy flight step size is used for fine-grained searching to ensure the accuracy of their optimal values; for insensitive parameters, a larger Levy flight step size is used for efficient exploration, avoiding unnecessary computational overhead. Simultaneously, initializing the sparrow population with typical slope instability modes allows the optimization process to start from a more physically meaningful region, further improving optimization efficiency and the reliability of the results. Through the above hyperparameter optimization processing method based on the SSA algorithm, the constructed SSA-XGBoost classification model can better perceive the non-convexity of the slope parameter space, thereby obtaining a better hyperparameter configuration, significantly improving the accuracy, robustness, and generalization ability of the slope stability assessment model, and making it more adaptable and reliable in complex and ever-changing slope engineering practices.

[0090] In some embodiments, this application proposes an improved slope stability assessment method, which further includes the following steps before using slope feature data samples as input features of a pre-defined SSA-XGBoost classification model:

[0091] Based on the slope feature data samples, the stability coefficient corresponding to each set of slope feature data samples is calculated using the preset limit equilibrium strip method, and the stability coefficient is used as the prior input feature of the SSA-XGBoost classification model.

[0092] Specifically, the limit equilibrium slice method is a classic slope stability analysis method. Its core idea is to divide the soil on the potential sliding surface of a slope into several vertical soil strips and calculate the stability coefficient of the entire sliding surface by analyzing the force balance of each strip. This method is based on rigorous mechanical equilibrium principles and can provide a stability evaluation index with clear physical meaning. During the calculation, it fully utilizes information such as slope geometry and geotechnical parameters contained in the slope feature data samples to perform a detailed force analysis on each soil strip, ultimately obtaining a numerical value representing the overall stability of the slope. Subsequently, the stability coefficient calculated by the limit equilibrium slice method is used as a priori input feature of the SSA-XGBoost classification model. This means that the stability coefficient, along with the original slope geometry, geotechnical parameters, and environmental dynamic parameters, is input into the SSA-XGBoost classification model for training. This approach of integrating traditional physical mechanics analysis results into a machine learning model aims to provide the model with additional, physically meaningful guidance information, thereby enhancing the model's physical constraints and interpretability.

[0093] Through the above technical solution, this application effectively integrates traditional physical and mechanical analysis results into a data-driven evaluation model. This approach provides the SSA-XGBoost classification model with strong physical constraints and prior knowledge, enabling the model to naturally conform to the principles of mechanical equilibrium when learning slope stability laws, thus avoiding the physical paradoxes that may arise from purely data-driven models. Therefore, the slope stability evaluation model obtained in this application not only possesses the learning capabilities of a data-driven model but also has physical rationality and interpretability, greatly enhancing the model's credibility and acceptability in practical engineering decisions and improving the reliability of the evaluation results.

[0094] Based on the above embodiments, this application further proposes that the limit equilibrium slice method includes: the simplified Bishop method or the unbalanced thrust method.

[0095] Specifically, the simplified Bishop method is a limit equilibrium method widely used in slope stability analysis, particularly suitable for circular slip surfaces. This method simplifies the force balance equations by dividing the potential sliding body into several vertical soil strips and assuming that the forces between the strips are horizontal. The overall stability coefficient of the slope is mainly solved through the moment balance equations. Due to its relatively simple calculation and reliable results, the simplified Bishop method has been widely used in engineering practice, especially for circular failure modes of homogeneous or layered soil slopes.

[0096] On the other hand, the unbalanced thrust method, such as the Janbu generalized slice method, is a more general limit equilibrium method capable of handling non-circular slip surfaces and more complex soil layer distributions and groundwater conditions. When considering the forces between soil slices, this method considers not only the horizontal component but also the vertical component and its inclination angle, solving for the stability coefficient by simultaneously satisfying force equilibrium and moment equilibrium conditions. The unbalanced thrust method can more accurately reflect the actual stress state and failure mechanism of complex slopes and is suitable for slope stability analysis with complex geological conditions and irregular slip surface shapes.

[0097] By employing the aforementioned technical solutions, the limit equilibrium slice method is specifically limited to either the simplified Bishop method or the unbalanced thrust method. This allows for the selection of the most suitable mechanical analysis method based on the actual geological conditions and potential failure modes of the slope when calculating the stability coefficients corresponding to slope feature data samples. Specifically, for slopes with circular sliding surface characteristics, the simplified Bishop method can be used for efficient and accurate stability coefficient calculation; while for slopes with complex geological structures and irregular sliding surface shapes, the unbalanced thrust method can be used for more refined and comprehensive stability analysis. This explicit limitation ensures that the stability coefficients, which serve as prior input features for the SSA-XGBoost classification model, have significantly improved physical meaning and computational accuracy. Consequently, the model's prediction results better conform to mechanical equilibrium constraints, avoiding the physical paradoxes that may arise from purely data-driven models, and greatly enhancing the model's credibility and acceptability in practical engineering decisions.

[0098] In some of the embodiments described above in this application, this application further proposes that, based on the lithology type corresponding to the slope feature data sample, when the lithology type similarity of any two sets of unstable samples is higher than a preset similarity threshold, the two sets of unstable samples are combined into a sample pair; interpolation is performed on each sample pair to generate a synthetic sample, which is used to expand the slope feature data sample.

[0099] Specifically, lithology is a crucial component of slope geological structure, directly influencing slope mechanical behavior and instability mechanisms. Using lithology as a basis for sample expansion ensures that the synthesized samples maintain physical consistency with the original samples, avoiding the generation of false data that does not reflect actual geological conditions. Lithology types can be obtained through geological survey reports, borehole data, geophysical exploration, etc., and can be classified and coded; for example, different lithologies such as sandstone, mudstone, and granite can be assigned different numerical values ​​or category labels.

[0100] This step aims to screen unstable samples with similar geological backgrounds to ensure the high physical plausibility of the subsequently interpolated synthetic samples. Lithological type similarity can be quantified in various ways. For example, for discrete lithological categories, it can be directly determined whether they belong to the same category; for lithological descriptions with continuous characteristics (such as mineral composition ratios), indices such as cosine similarity and the reciprocal of Euclidean distance can be used for calculation. The preset similarity threshold is set based on practical engineering experience and data characteristics to balance the quantity and quality of sample expansion. When the lithological type similarity of two sets of unstable samples reaches or exceeds this threshold, it indicates that they have sufficient commonality in key geological attributes and are suitable for constructing sample pairs.

[0101] Interpolation is a core technique for generating synthetic samples. For existing sample pairs, various interpolation algorithms can be used to generate new samples that fall between the two. For example, linear interpolation can be used, which involves a weighted average of the feature parameters in the sample pair; the weights can be randomly generated or determined according to specific rules. More complex interpolation methods can also be used, such as a variant of the SMOTE (Synthetic Minority Over-sampling Technique) algorithm, which generates new synthetic samples in the feature space along the line segments connecting minority class samples. Through interpolation, new, physically meaningful sample points can be created in the feature space of existing unstable samples, effectively increasing the number of unstable samples.

[0102] The generated synthetic samples were added to the original slope feature data sample set. This augmentation not only increased the number of instability samples, alleviating the sample imbalance problem, but also, because the synthetic samples were generated by interpolation based on the lithological similarity of real instability samples, their distribution in the feature space is reasonable, better reflecting the inherent laws of slope instability. The augmented dataset will be used to train the SSA-XGBoost classification model, providing the model with richer and more comprehensive learning materials.

[0103] Through the above technical solution, this application effectively solves the problem of scarce slope instability samples. By identifying instability samples with similar lithological types and interpolating them to generate synthetic samples, the model can be exposed to more diverse instability modes during training, thereby enhancing its learning ability in real physical space. This sample expansion method based on geological similarity ensures the physical rationality of the synthetic samples and avoids blindly generating invalid data that may lead to model misjudgment. Therefore, the SSA-XGBoost classification model can learn instability characteristics more fully during training, improving the generalization ability of the slope stability assessment model and the accuracy of identifying potential instability risks, especially in the face of complex and variable geological conditions, providing more reliable assessment results.

[0104] The above is a detailed description of the key technical steps of an embodiment of a slope stability assessment method provided in this application. The following is a more specific example to illustrate the above technical solution in more detail:

[0105] Assuming a scenario where a power company is constructing a transmission line in a mountainous area, and needs to conduct stability assessments on multiple slopes along the line, the implementation process of the slope stability assessment method provided in this application can be seen in the following example:

[0106] like Figure 2 As shown, firstly, historical slope data and on-site survey data along the power facility route were collected to construct slope characteristic data samples. Each set of slope characteristic data samples includes: slope geometry, geotechnical parameters, and environmental dynamic parameters. Specifically, geometric features such as slope height, slope angle, and slope curvature were obtained through 3D laser scanning technology; geotechnical parameters such as cohesion, internal friction angle, unit weight, and permeability coefficient were obtained by combining in-situ direct shear tests and indoor triaxial tests, and a geotechnical mechanical response database was established; simultaneously, an IoT sensor network was deployed to collect environmental dynamic factors such as groundwater level, rainfall, and seismic motion parameters in real time. These data form the basis for model training.

[0107] Before model training, the evaluation team performed preprocessing to enhance the model's physical constraints and address the scarcity of instability samples. Firstly, for each set of slope feature data samples, a simplified Bishop method was used to calculate its corresponding stability coefficient, which was then used as prior input features for the SSA-XGBoost classification model. This approach ensures the model naturally conforms to mechanical equilibrium constraints during prediction, avoiding potential physical paradoxes that might arise from purely data-driven models, significantly improving the model's credibility in engineering decisions. Secondly, to address the scarcity of instability samples, the lithology of the slope feature data samples was used to detect whether any two sets of instability samples had a lithology type similarity exceeding a preset similarity threshold. If so, these two sets of instability samples were grouped into a sample pair, and interpolation was performed on this pair to generate a synthetic sample, thereby expanding the slope feature data sample set and enhancing the model's learning ability in real physical space. In some embodiments, after obtaining new samples, the "theoretical safety factor" of the new samples can be calculated using the simplified Bishop method or the unbalanced thrust method to verify their physical rationality. To avoid generating "pseudo-samples" that are physically impossible or violate geological laws, and to enhance the model's learning ability in real physical space.

[0108] Next, the SSA-XGBoost classification model was constructed and trained. The model construction process fully considered the non-convex nature of the slope parameter space. Specific steps included: For each feature parameter included in the slope feature data samples, a global sensitivity analysis was performed using the Morris screening method to determine the sensitivity of each feature parameter to slope stability, and the feature parameters were categorized into sensitive and insensitive parameters based on their sensitivity. For example, slope angle and cohesion might be identified as sensitive parameters, while some minor geometric features might be identified as insensitive parameters. Subsequently, based on the sensitivity of each feature parameter and the pre-defined correspondence between parameter sensitivity and Levy flight step size, the corresponding Levy flight step size for each feature parameter was determined. The Levy flight step size for sensitive parameters was smaller than that for insensitive parameters, allowing for a more refined search of key parameters during the optimization process. Then, based on the slope feature data samples and according to the typical parameter ranges corresponding to typical slope instability modes, multiple sparrow populations were initialized. Finally, based on the sparrow population and the correspondence between feature parameters and Levy flight stride length, the initial XGBoost classification model was hyperparameter optimized according to the optimization logic of the SSA optimization algorithm, resulting in an SSA-XGBoost classification model. This SSA hyperparameter optimization for non-convex slope spaces allows the model to better perceive the complex characteristics of the slope parameter space, leading to a global optimization algorithm.

[0109] During the model training phase, preprocessed slope feature data samples were used as input features for the SSA-XGBoost classification model. The model training incorporated the built-in geological weighted loss function within the SSA-XGBoost classification model. This geological weighted loss function is a weighted function constructed based on geological complexity weights and the FocalLoss function. Geological complexity weights quantify the geological structural complexity of the slope samples; for example, conditions with developed faults and complex, variable lithology are assigned higher geological complexity weights. The FocalLoss function effectively addresses the class imbalance problem, allowing the model to focus more on difficult-to-classify samples, especially unstable samples, during training. By combining geological complexity weights and the FocalLoss function, the model can more accurately assess slope stability under different geological conditions and effectively avoid the majority class bias problem of traditional XGBoost models when facing imbalanced data, thus improving the sensitivity for identifying potentially dangerous slopes.

[0110] During training, the system determines the convergence degree of the SSA-XGBoost classification model based on the calculation results of the geological weighted loss function. When the convergence degree or the number of iterations meets the preset training termination condition, the training process stops, and the final slope stability assessment model is obtained.

[0111] After the model training is completed, before it is officially put into practical application, a stratified cross-validation method can be used to validate the trained model. Specifically, the dataset is divided into layers according to the slope lithology type (such as soil, rock, and composite type) to ensure that each lithology sample is evenly distributed in the training set and the validation set. If the validation effect does not meet the requirements, the training process can be restarted.

[0112] When a slope stability assessment of a target area is required, the slope characteristic data of that area is first acquired, including its geometric structure, geotechnical parameters, and environmental dynamic parameters. This data is then input into a pre-trained slope stability assessment model. Before the model calculates the slope stability assessment result for the target area, the system also monitors the rate of change of environmental dynamic parameters in real time. For example, when the rate of change of rainfall or groundwater level is detected to be greater than a preset threshold, the system dynamically increases the focused parameters in the geological weighted loss function based on the preset correspondence between parameter change rates and focused parameters. Conversely, when the rate of change is not greater than the preset threshold, the focused parameters are decreased. This "dynamic weight adjustment based on monitoring data stream" mechanism allows the model to adjust its sensitivity to boundary samples according to real-time rainfall and water level data, resulting in faster response speed and higher early warning timeliness in early warning scenarios, effectively avoiding the problems of delayed response and untimely early warning in traditional methods. Finally, through the calculation of the slope stability assessment model, the slope stability assessment result for the target area is obtained, providing a scientific basis for the operation and maintenance and risk management of power facilities.

[0113] The above is a detailed description of an embodiment of a slope stability assessment method provided in this application. The following is a detailed description of related embodiments of a slope stability assessment device, terminal, and storage medium provided in this application.

[0114] Please see Figure 3 This application proposes a slope stability assessment device for performing the above-described slope stability assessment method, the main components of which include:

[0115] The sample acquisition unit 201 is used to acquire slope feature data samples, wherein each set of slope feature data samples includes: slope geometry, soil and rock parameters and environmental dynamic parameters.

[0116] The evaluation model training unit 202 is used to train the model based on the preset SSA-XGBoost classification model, taking the slope feature data samples as input features of the SSA-XGBoost classification model, and combining the geological weighted loss function built into the SSA-XGBoost classification model. The geological weighted loss function is a weighted function constructed based on geological complexity weights and FocalLoss function. The geological complexity weights are used to quantify the geological structural complexity of the slope samples.

[0117] The model convergence determination unit 203 is used to determine the convergence degree of the SSA-XGBoost classification model based on the calculation result of the geological weighted loss function. When the convergence degree or the number of iterations meets the preset training termination condition, the slope stability assessment model is obtained.

[0118] The slope stability measurement and evaluation unit 204 is used to acquire slope characteristic data of the target area, input the slope characteristic data into the slope stability evaluation model, and obtain the slope stability evaluation result of the target area through the calculation of the slope stability evaluation model.

[0119] This embodiment constructs a geological weighted loss function by combining geological complexity weights with the FocalLoss function in a weighted manner, and integrates this loss function into the evaluation model training unit of the slope stability assessment device. This allows the model training process to simultaneously consider the complexity of geological structures and the imbalance of sample categories, thereby improving the sensitivity of identifying potentially dangerous slopes and effectively reducing the missed detection of instability risks.

[0120] Through the above technical solution, the device achieves efficient and accurate assessment of slope stability. Especially in engineering scenarios where instability samples are scarce, it significantly improves the model's ability to identify slopes with complex geological structures, avoiding the shortcomings of low computational efficiency of traditional methods and high risk of missed detection in existing machine learning models, and providing reliable technical support for the safe operation of power facilities.

[0121] Preferably, before obtaining the slope stability assessment result of the target area through the calculation of the slope stability assessment model, the method further includes:

[0122] The rate of change of the environmental dynamic parameters is detected, and the focus parameters in the geological weighted loss function are dynamically updated based on the preset correspondence between the parameter change rate and the focus parameters.

[0123] Furthermore, the step of detecting the rate of change of the environmental dynamic parameters and dynamically updating the focus parameters in the geological weighted loss function by combining the preset correspondence between the parameter change rate and the focus parameters includes:

[0124] When the rate of change of the environmental dynamic parameters is greater than the preset rate of change threshold, the focus parameter in the geological weighted loss function is increased according to the preset correspondence between the parameter rate of change and the focus parameter. When the rate of change of the environmental dynamic parameters is not greater than the preset rate of change threshold, the focus parameter in the geological weighted loss function is decreased according to the preset correspondence between the parameter rate of change and the focus parameter.

[0125] Furthermore, the construction method of the SSA-XGBoost classification model includes:

[0126] Based on each feature parameter included in the slope feature data sample, a global sensitivity analysis is performed on the feature parameter using the Morris screening method or the Sobol index method to determine the sensitivity of each feature parameter to slope stability, and based on the sensitivity, the feature parameter is divided into sensitive parameters and insensitive parameters.

[0127] Based on the sensitivity corresponding to each feature parameter, and combined with the preset correspondence between parameter sensitivity and Levy flight step size, the Levy flight step size corresponding to each feature parameter is determined, wherein the Levy flight step size corresponding to the sensitive parameter is smaller than the Levy flight step size corresponding to the non-sensitive parameter.

[0128] Based on the slope characteristic data sample, the sparrow population is initialized according to the typical parameter range corresponding to the typical slope instability mode;

[0129] Based on the sparrow population and the correspondence between feature parameters and Levy flight stride length, the initial XGBoost classification model is hyperparameter optimized according to the optimization logic of the SSA optimization algorithm, resulting in the SSA-XGBoost classification model.

[0130] Furthermore, before using the slope feature data samples as input features of the SSA-XGBoost classification model based on the preset SSA-XGBoost classification model, the method further includes:

[0131] Based on the slope feature data samples, the stability coefficient corresponding to each set of slope feature data samples is calculated using the preset limit equilibrium strip method, so as to use the stability coefficient as the prior input feature of the SSA-XGBoost classification model.

[0132] Furthermore, the limit equilibrium slice method includes: the simplified Bishop method or the unbalanced thrust method.

[0133] Furthermore, before using the slope feature data samples as input features of the SSA-XGBoost classification model based on the preset SSA-XGBoost classification model, the method further includes:

[0134] Based on the lithology type corresponding to the slope feature data samples, when the lithology type similarity of any two sets of unstable samples is higher than a preset similarity threshold, the two sets of unstable samples are combined into a sample pair.

[0135] Interpolation is performed on each sample pair to generate a synthetic sample, which is used to expand the slope feature data sample.

[0136] In some embodiments, this application also discloses a slope stability assessment terminal, including: a memory and a processor; the memory is used to store program code corresponding to the above-described slope stability assessment method; the processor is used to read and execute the program code to implement the slope stability assessment method.

[0137] In practical implementation, the slope stability assessment terminal is configured as a dedicated embedded device or a general-purpose computing platform. The memory is configured as a non-volatile storage medium, including random access memory and flash memory, to store the program code implementing the aforementioned slope stability assessment method. This program code is configured to include a sequence of instructions for acquiring slope feature data samples, training the model based on the SSA-XGBoost classification model, determining the model convergence degree, and assessing the slope stability of the target area. The processor is configured as a central processing unit or graphics processing unit to read and execute the program code, thereby achieving real-time processing and stability assessment of the slope feature data. Specifically, the terminal connects to an IoT sensor network via a communication interface to receive dynamic environmental parameter data in real time; the pre-stored program code in the memory is configured to call the SSA-XGBoost classification model, performing calculations using a geologically weighted loss function constructed by combining geological complexity weights and the FocalLoss function; the processor is configured to determine the model convergence state based on the calculation results of the geologically weighted loss function and output the slope stability assessment results when the training termination condition is met.

[0138] This embodiment combines memory and processor in a hardware-coordinated manner, thereby solidifying the aforementioned slope stability assessment method into executable program code, achieving real-time and efficient assessment of slope instability risk. Specifically, the terminal is configured to be deployed at power facility sites or monitoring centers. It automatically executes the acquisition of slope feature data samples, training of the SSA-XGBoost classification model, and target area assessment process through pre-stored program code in memory. This effectively solves the problem of missed detection of instability risks due to class imbalance, improves the sensitivity of identifying slopes with complex geological structures, and reduces computational resource consumption. Thus, it achieves the technical effect of timely early warning of slope instability risks and avoiding missed detections in power facility operation and maintenance.

[0139] In some embodiments, this application also discloses a computer-readable storage medium storing program code that is read and executed by a processor to implement the above-described slope stability assessment method.

[0140] When the program code stored in the storage medium is executed, it can drive the processor to make a comprehensive evaluation based on the slope geometry, soil and rock parameters and environmental dynamic parameters. The geological complexity weight is used to quantify the geological complexity of the slope sample, while the FocalLoss function dynamically adjusts the focusing parameters to make the model pay more attention to unstable samples that are difficult to classify during training, especially enhancing the sensitivity to boundary samples in slope scenarios with complex geological conditions.

[0141] Through the above technical solution, slope stability assessment can efficiently process multi-source slope characteristic data, improving the early warning capability for dangerous slopes while ensuring computational efficiency. This solution not only inherits the global optimization capability of the SSA-XGBoost classification model for non-convex parameter spaces, but also ensures the model's accuracy in identifying low-probability instability events in typical power facility application scenarios through the synergistic effect of geological complexity weights and the FocalLoss function, thus providing reliable technical support for the safe operation of the power grid.

[0142] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the terminals, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0143] In the several embodiments provided in this application, it should be understood that the disclosed terminals, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0144] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0145] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0146] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0147] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0148] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0149] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for assessing slope stability, characterized in that, include: Obtain slope feature data samples, where each set of slope feature data samples includes: slope geometry, geotechnical parameters, and environmental dynamic parameters; Based on the preset SSA-XGBoost classification model, the slope feature data samples are used as input features of the SSA-XGBoost classification model. The model is trained by combining the geological weighted loss function built into the SSA-XGBoost classification model. The geological weighted loss function is a weighted function constructed based on geological complexity weights and FocalLoss function. The geological complexity weights are used to quantify the geological structural complexity of the slope samples. Based on the calculation results of the geological weighted loss function, the convergence degree of the SSA-XGBoost classification model is determined. When the convergence degree or the number of iterations meets the preset training termination condition, the slope stability assessment model is obtained. Obtain slope characteristic data of the target area, input the slope characteristic data into the slope stability assessment model, and obtain the slope stability assessment result of the target area through the calculation of the slope stability assessment model.

2. The slope stability assessment method according to claim 1, characterized in that, Before obtaining the slope stability assessment results for the target area through the calculation of the slope stability assessment model, the following steps are also included: The rate of change of the environmental dynamic parameters is detected, and the focus parameters in the geological weighted loss function are dynamically updated based on the preset correspondence between the parameter change rate and the focus parameters.

3. The slope stability assessment method according to claim 2, characterized in that, The process of detecting the rate of change of the environmental dynamic parameters and dynamically updating the focused parameters in the geological weighted loss function, based on the preset correspondence between the parameter change rates and the focused parameters, includes: When the rate of change of the environmental dynamic parameters is greater than the preset rate of change threshold, the focus parameter in the geological weighted loss function is increased according to the preset correspondence between the parameter rate of change and the focus parameter. When the rate of change of the environmental dynamic parameters is not greater than the preset rate of change threshold, the focus parameter in the geological weighted loss function is decreased according to the preset correspondence between the parameter rate of change and the focus parameter.

4. The slope stability assessment method according to claim 1, characterized in that, The construction method of the SSA-XGBoost classification model includes: Based on each feature parameter included in the slope feature data sample, a global sensitivity analysis is performed on the feature parameter using the Morris screening method or the Sobol index method to determine the sensitivity of each feature parameter to slope stability, and based on the sensitivity, the feature parameter is divided into sensitive parameters and insensitive parameters. Based on the sensitivity corresponding to each feature parameter, and combined with the preset correspondence between parameter sensitivity and Levy flight step size, the Levy flight step size corresponding to each feature parameter is determined, wherein the Levy flight step size corresponding to the sensitive parameter is smaller than the Levy flight step size corresponding to the non-sensitive parameter. Based on the slope characteristic data sample, the sparrow population is initialized according to the typical parameter range corresponding to the typical slope instability mode; Based on the sparrow population and the correspondence between feature parameters and Levy flight stride length, the initial XGBoost classification model is hyperparameter optimized according to the optimization logic of the SSA optimization algorithm, resulting in the SSA-XGBoost classification model.

5. The slope stability assessment method according to claim 1, characterized in that, The method based on the preset SSA-XGBoost classification model further includes the following steps before using the slope feature data samples as input features of the SSA-XGBoost classification model: Based on the slope feature data samples, the stability coefficient corresponding to each set of slope feature data samples is calculated using the preset limit equilibrium strip method, so as to use the stability coefficient as the prior input feature of the SSA-XGBoost classification model.

6. The slope stability assessment method according to claim 5, characterized in that, The limit equilibrium slice method includes: the simplified Bishop method or the unbalanced thrust method.

7. The slope stability assessment method according to claim 1, characterized in that, The method based on the preset SSA-XGBoost classification model further includes the following steps before using the slope feature data samples as input features of the SSA-XGBoost classification model: Based on the lithology type corresponding to the slope feature data samples, when the lithology type similarity of any two sets of unstable samples is higher than a preset similarity threshold, the two sets of unstable samples are combined into a sample pair. Interpolation is performed on each sample pair to generate a synthetic sample, which is used to expand the slope feature data sample.

8. A slope stability assessment device, characterized in that, A method for performing slope stability assessment as described in any one of claims 1 to 7, comprising: The sample acquisition unit is used to acquire slope feature data samples. Each set of slope feature data samples includes: slope geometry, soil and rock parameters, and environmental dynamic parameters. The evaluation model training unit is used to train the model based on the preset SSA-XGBoost classification model, taking the slope feature data samples as input features of the SSA-XGBoost classification model, and combining the geological weighted loss function built into the SSA-XGBoost classification model. The geological weighted loss function is a weighted function constructed based on geological complexity weights and the FocalLoss function. The geological complexity weights are used to quantify the geological structural complexity of the slope samples. The model convergence determination unit is used to determine the convergence degree of the SSA-XGBoost classification model based on the calculation results of the geological weighted loss function. When the convergence degree or the number of iterations meets the preset training termination condition, the slope stability assessment model is obtained. The slope stability measurement and evaluation unit is used to acquire slope characteristic data of the target area, input the slope characteristic data into the slope stability evaluation model, and obtain the slope stability evaluation result of the target area through the calculation of the slope stability evaluation model.

9. A slope stability assessment terminal, characterized in that, include: Memory and processor; The memory is used to store program code, which corresponds to the slope stability assessment method as described in any one of claims 1 to 7; The processor is used to read and execute the program code to implement the slope stability assessment method.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that is read and executed by a processor to implement the slope stability assessment method as described in any one of claims 1 to 7.